CN-122014539-A - Wind blade defect detection system based on event camera driving

CN122014539ACN 122014539 ACN122014539 ACN 122014539ACN-122014539-A

Abstract

The invention discloses a wind blade defect detection system based on event camera driving, and belongs to the technical field of wind power generation equipment state monitoring and intelligent detection. The system comprises an unmanned aerial vehicle control unit, an event sensing unit, a data processing unit, a defect detection unit and a result output unit. And cruising the unmanned aerial vehicle carrying the event camera along a preset route, collecting visual event streams on the surface of the blade, and realizing time alignment with IMU data by means of a hardware synchronization mode. The system carries out filtering, motion compensation and feature construction on the event stream, generates an event feature map, then inputs the event feature map into a neural network model, completes classification, positioning and three-dimensional coordinate mapping of defects, and finally outputs a visual report containing defect multidimensional information. The method effectively overcomes the limitation of the traditional visual detection under high-speed movement and complex illumination, and realizes efficient and accurate identification and quantitative evaluation of the defects of the wind turbine blade.

Inventors

NIU XIAOTONG
WU WEI
ZHENG HAIFENG
ZHUANG JUN

Assignees

上海中认尚科新能源技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (12)

1. A wind blade defect detection system based on event camera driving, comprising: the unmanned aerial vehicle control unit is used for cruising based on a preset spiral route through the unmanned aerial vehicle positioning module and the laser radar obstacle avoidance module; The system comprises an event sensing unit, an event processing unit and a control unit, wherein the event sensing unit comprises an event camera module, an IMU inertial measurement module and a synchronization module, the event camera module acquires a blade surface visual event stream through a dynamic visual sensor, the IMU inertial measurement module acquires inertial motion data of the unmanned aerial vehicle, and the synchronization module realizes time synchronization of the event stream and the IMU data through a hardware triggering mode; The data processing unit is used for filtering, motion compensation and feature construction of the event stream and outputting an event feature map; The defect detection unit is used for carrying out defect classification and positioning on the event feature map, and processing the event feature map by utilizing a neural network model to obtain the type, the confidence coefficient and the two-dimensional boundary frame of the defect; and the result output unit is used for displaying the detection result in real time and mapping the two-dimensional position of the defect to a three-dimensional coordinate system of the blade to obtain three-dimensional coordinates and size information and generate a visual report containing multi-dimensional information of the defect.
2. The system of claim 1, wherein the event sensing unit further comprises an adaptive laser light supplementing module for dynamically adjusting power according to event density to maintain integrity of event features in low light environments.
3. The wind blade defect detection system based on event camera driving according to claim 1, wherein the motion compensation of the data processing unit performs rotation and translation transformation on event coordinates based on gesture data provided by the IMU inertial measurement module to compensate for image offset caused by unmanned aerial vehicle motion.
4. An event camera-driven wind blade defect detection system according to claim 1, wherein the motion compensation of the data processing unit is further adapted to periodically perform a cumulative error calibration based on fixed characteristics of the wind blade surface.
5. The wind blade defect detection system based on event camera driving according to claim 1, wherein the feature construction of the data processing unit accumulates and converts asynchronous event streams into a multi-channel event feature map using a dynamic time window strategy, the event feature map comprising time surface, polarity histogram and event density information.
6. An event camera-driven wind blade defect detection system according to claim 1, wherein the defect detection unit employs a neural network model comprising event feature attention mechanisms, enhancing defect features and suppressing background features.
7. The system of claim 1, wherein the synchronization module triggers the modulated light source to excite brightness fluctuation to generate an event stream for active excitation when the event density of the static area is detected to be continuously lower than a threshold value, so as to enhance the event response of the static defect.
8. The event camera-driven wind blade defect detection system of claim 1 wherein the result output module further comprises predicting a trend of increase in defect size based on historical detection data and outputting a staged repair recommendation.
9. The system of claim 1, wherein the event profile comprises a multi-information channel including at least a time surface channel with a most recent event time stamp as a pixel value, a polarity histogram channel that records positive and negative polarity event counts, and an event density map channel that reflects the number of events per unit area.
10. The system for detecting defects of wind turbine blades based on event camera driving according to claim 1, wherein the hardware triggering mode is as follows: And outputting a periodic hardware pulse signal to the synchronization module by the IMU inertial measurement module, wherein the synchronization module marks a uniform time stamp on each event acquired by the event camera module based on the pulse signal.
11. An event camera-driven wind blade defect detection system according to claim 6 wherein the event feature attention mechanism comprises: In the characteristic extraction stage of the network, calculating and inputting an event density map of each candidate region on the characteristic map in real time, wherein the event density map is obtained based on an event density information channel generated by the data processing unit; For the space position on any feature map, the attention weight is dynamically determined by the event density corresponding to the position and the average event density of the neighborhood background; Multiplying the calculated attention weight and the original feature map element by element in the channel dimension to realize the enhancement of the defect high-response region features and the suppression of the background low-response region features; Parameters of the event feature attention mechanism are trained together with the neural network model, and the optimization target is to introduce weight sparsity regular terms into the classification and positioning loss functions so as to enable the network to learn to focus on the region with remarkable abnormal event density in the training process.
12. The system for detecting defects of wind turbine blades based on event camera driving according to claim 1, wherein the visual report generated by the result output unit further comprises a structured defect list generated according to standard operation and maintenance specifications of wind turbine blades, and the list at least comprises a blade number, a detection time, a defect type, three-dimensional position coordinates, an equivalent two-dimensional size, an estimated depth, a confidence level and a recommended maintenance priority.

Description

Wind blade defect detection system based on event camera driving Technical Field The invention belongs to the technical field of wind power generation equipment state monitoring and intelligent detection, and particularly relates to a wind blade defect detection system driven by an event camera. Background The wind driven generator blade is a core component for capturing wind energy, and the structural integrity of the wind driven generator blade directly determines the generating efficiency and the operation safety of the unit. The blade is exposed to outdoor extreme environment for a long time, bears the effects of sand erosion, alternating load, ultraviolet aging, strong light irradiation and the like, and is easy to generate the defects of surface cracks, local corrosion, gel coat falling and the like. According to industry statistics, the single economic loss can reach millions of yuan due to failure accidents caused by failure of timely detection of blade defects, and the blade faults account for more than 30% of the total downtime of a wind power plant, so that the failure detection method is a key bottleneck for restricting the reliability of wind power equipment. The existing blade defect detection technology mainly depends on a traditional visual sensor (a frame exposure camera) or single-mode equipment, and has the core limitations that when the traditional frame exposure camera is used for unmanned aerial vehicle high-speed inspection, motion blur is easily generated due to unmatched exposure time and motion speed, so that 1 mm-level micro cracks are missed to be detected, the dynamic range of the traditional frame exposure camera cannot cover the brightness difference of a strong light reflection area-shadow area of the surface of a blade, and the defect characteristics are completely submerged due to overexposure or underexposure. Although three-dimensional sensing equipment such as laser radar and the like can acquire depth information, the characteristic identification degree of static defects (such as stable corrosion areas) is low, real defects and surface stains cannot be distinguished, manual secondary interpretation is needed, single-blade detection takes more than 60 minutes, and the efficiency is low. In the prior art, the fusion of vision and three-dimensional data stays at a simple splicing level, and a dynamic association mechanism aiming at defect characteristics is lacking, so that the accuracy deviation of two-dimensional positioning and three-dimensional quantization is caused, and the maintenance decision is difficult to support. The event camera is used as a novel biological heuristic sensor, and only outputs an asynchronous event stream (comprising space coordinates, time stamps and brightness change polarities) when the brightness of pixels changes by a preset threshold value, so that the novel biological heuristic sensor has the advantages of high dynamic range, microsecond time resolution, no motion blur, low data redundancy and the like, and provides a new paradigm for solving the defect detection problem in a complex environment. However, the event camera outputs unstructured event streams, and the feature extraction, the dynamic association with three-dimensional data and the long-time tracking mechanism do not form a mature scheme, and in particular, in curved dynamic scenes such as wind blades, the mapping relation between event features and defect physical properties still needs to be broken through. Therefore, the research and development of the defect detection system and method based on event camera driving, which combines virtual-real fusion sensing, dynamic characteristic enhancement and multi-mode fusion verification technology, breaks through the bottleneck of environmental adaptability and quantization accuracy of traditional detection, and becomes a key technical problem to be solved in the field of wind power generation operation and maintenance. Disclosure of Invention In order to solve the problems in the prior art, the invention provides a wind blade defect detection system driven by an event camera, and the aim of the invention can be achieved by the following technical scheme: A wind blade defect detection system based on event camera driving, comprising: the unmanned aerial vehicle control unit is used for cruising based on a preset spiral route through the unmanned aerial vehicle positioning module and the laser radar obstacle avoidance module; The system comprises an event sensing unit, an event processing unit and a control unit, wherein the event sensing unit comprises an event camera module, an IMU inertial measurement module and a synchronization module, the event camera module acquires a blade surface visual event stream through a dynamic visual sensor, the IMU inertial measurement module acquires inertial motion data of the unmanned aerial vehicle, and the synchronization module realizes time synchronization of the event stream and the IMU da